With General Motors' transition to the cloud, much of the company's data began migrating from local servers to a mix of cloud providers. Our team needed to understand the challenges of that transition and build a tool to help people find the data they needed. I worked alongside two other UX designers, and led the project from ideation through the final high-fidelity prototype.
User problems
- No centralized data catalog for all of GM. Data lived on local servers or across providers like AWS and Azure, with no single index.
- Unclear limitations and resources. There was little documentation on requesting access, subscriptions, or repos.
- No coding standards or cost structure. This left a lot of room for error, driving up costs and triggering warnings.
Solution
A central hub for data exploration — documenting available datasets and metadata like location, descriptions, lineage, and usage instructions, alongside clear onboarding guidance for cloud providers and coding standards for cost efficiency. Together, these features were built to save people time and keep development costs down.
User interviews
The primary users were data scientists and developers at GM. I wrote a script for semi-structured interviews to understand which data tools they used and how — 13 participants across 4 user types, recruited directly from target organizations.
Results
Getting access to data could take months, sending people in circles to find an approver. Teams moving to the cloud weren't given the support or guidelines they needed, and were still ramping up costs without proper flows in place for resources like CPU.
I organized the interview notes into an affinity map by likes, dislikes, and needs, which made the findings far easier to act on. One surprise: most users were already on Databricks despite the recent push to the cloud — I'd assumed more people would still be on-prem.
Personas
From the affinity mapping, I distilled the research into three personas that guided the rest of the design process.
Assumption mapping
GM as a whole still lacked information about working with cloud data, so I mapped out where we were making assumptions — helping us decide which features were most desirable and feasible, and where we needed to collaborate with other teams to fill gaps.
Stakeholder workshop
Before continuing, we confirmed the vision, goals, and high-level features with stakeholders. I helped prepare and lead a two-day workshop — organizing an agenda around Data Discovery, Data Marketplace, SQLGPT & existing products, and future products — with How-Might-We questions for each topic and sticky-note responses from the full team, including our director.
User flows
With required features finalized, I mapped current- and future-state user flows for each persona and use case, marking key steps in purple and open questions in red. This snippet shows one of the more extensive future-state flows — the full set is on Figma.
Inspiration & sketching
I looked at comparable products to inform direction, then moved into Crazy 8's sketching once the research, requirements, and flows were settled.
Information architecture
Sketching revealed we still lacked clarity on metadata for the data-item page. I met with stakeholders across adjacent teams to lock down the requirements, then organized an IA chart to visualize the final structure.
Low fidelity
Wireframes focused on short navigation paths and clear search/filtering. Teammates voted on the directions to carry into high fidelity.
Home page — one option with extra top tabs, one with a side nav.
Data search page — one grouped by category with suggested searches, one with filters and a flat list.
Data item page — three options varying the order of information and features like use cases and sample queries.
High fidelity
We settled on a light, subtle theme — appropriate for the density of information and quietly reinforcing GM's brand through a blue palette without overwhelming users with color.
After finalizing the prototype, I ran 4 usability tests with users from the original interviews — feedback on filters and metadata organization is already reflected below. Using auto-layout, components, and Dev Mode kept handoff to engineering fast.
Home page — search and filters, with the ability to save filter sets and clear them, plus a domains dropdown for browsing by group.
Search page — a toggleable filter panel and sorting by popularity, date added, or alphabetical order.
Data item page — description, data profile, and lineage, with collapsible sections so owners can edit metadata without cluttering the page for everyone else.
.png)
Conclusion
The Data Catalog is currently in development, with v1 soon to launch. It was the first project I carried from idea to development at GM — I wasn't able to see it ship, but I presented the final designs at our org's all-hands, and users were excited about the time the search and item-page information would save them.
Getting the information and feedback I needed despite communication barriers and siloing across a large IT organization meant staying close to my manager and PM through standups and extra syncs. I came in without cloud-space knowledge, so a lot of this project was building that technical fluency alongside the design work. If I could change one thing, I'd have run usability tests earlier in the process.